Problem

Not all Kubernetes metrics are present in StackState.

Cause

The StackState Agent uses two different sources to retrieve metrics from a Kubernetes cluster:

• The Kubelet API
• A kube-state-metrics pod

If an existing kube-state-metrics deployment is used, it is possible that this will not be auto-discovered by the StackState Agent. The auto-discovery mechanism used by the Agent relies on the image name used by the Pod. If this is not named kube-state-metrics, the Pod will not be auto-discovered and a number of metrics will be missing in StackState.

Solution

The image name used by the kube-state-metrics Pod should be manually added to the values.yaml file used for your Helm deployment. This will override the default identifier used by the StackState Agent to auto-discover the kube-state-metrics Pod. 

Add the following custom values to the values.yaml file used for your Helm deployment:

agent:
  config:
    override:
      - name: auto_conf.yaml
        path: /etc/stackstate-agent/conf.d/kubernetes_state.d
        data: |
          ad_identifiers:
            - my-custom-kubestatemetrics-image

          init_config:

          instances:
            - kube_state_url: http://%%host%%:8080/metrics

In this case, the Pod is using an image called my-custom-kubestatemetrics-image and will be auto-discovered correctly after deployment of the StackState Agent Helm chart.